recentpopularlog in

mcherm : statistics   86

« earlier  
How many atheists are there? - Marginal REVOLUTION
Surveys show 3%-11% atheists in the US. This researcher accounts for underreporting and gets 26%.
atheism  sociology  via:reddit  statistics 
may 2018 by mcherm
CMV: The Wilson effect definitively proves that intelligence is about 80% hereditary, and there is no more debate as to whether heredity or environmental influence plays a greater role.
A really good, understandable explanation of why measured heritability is not the same as amount by which the genes control things. Succinctly, "the environment can have more or less effect depending on how much it varies". Also, adoptive twins skew to higher incomes so twin studies can be misleading.
science  statistics  via:reddit  ScientificMethod 
february 2018 by mcherm
A.I. ‘Bias’ Doesn’t Mean What Journalists Say it Means – Jacobite
Many journalists write scare stories about biased AI algorithms when actually the algorithm is race-blind but actual outcomes are not. This may be true, but it also misses some important subtleties.
ai  racism  statistics  via:SlateStarCodex  journalism 
january 2018 by mcherm
Violence in Blue
1/10 of US killings is from police; 1/3 of all killings by a stranger is by police. This estimate is made by comparing incomplete lists to estimate population size and figuring in list correlation.
statistics  police  via:HackerNews  policeabuse 
december 2017 by mcherm
Same Stats, Different Graphs: Generating Datasets with Varied Appearance and Identical Statistics through Simulated Annealing | Autodesk Research
You can have VERY different visual appearances for the same mean, standard deviation, and other summary statistics. This paper has examples, and explains the technique they used to generate the data sets.
statistics  datavisualization  via:boingboing  math 
may 2017 by mcherm
How ZIP codes nearly masked the lead problem in Flint
The boundaries of the water system match the city boundary, but NOT the zip code boundaries. Actually, very little matches zip code boundaries.
statistics  personal_net  via:HackerNews 
may 2017 by mcherm
US Death Penalty Wrongful Convictions Executions | Time.com
Secondhand report on research claiming about 4% of people on death row are innocent.
prison  law  ethics  via:PaulGraham  science  statistics 
march 2017 by mcherm
Islamophobic Myths Debunked : islam
Muslims don't commit huge amounts of terrorism. This collection of stats helps show that. (Of course, you have to ignore WARS to get these numbers, but it is not unfair to claim war and terrorism are different things.)
terrorism  via:reddit  statistics 
december 2016 by mcherm
These Charts Show Why Minimum Wage Hikes Really Work - CityLab
Actual statistics showing that raising the minimum wage tends to increase salaries without decreasing employment.
statistics  economics  politics 
december 2016 by mcherm
The Pyramid And The Garden | Slate Star Codex
Using numerology to understand the risks of trusting scientific studies based solely on P-value.
ScientificMethod  statistics  SlateStarCodex  via:reddit 
november 2016 by mcherm
How One 19-Year-Old Illinois Man Is Distorting National Polling Averages - The New York Times
How a poll overweight things and ends up with one person swinging the results by a notable amount. None of this understanding would be possible without wonderfully transparent data sharing by the pollsters.
statistics  nytimes  via:reddit 
october 2016 by mcherm
Beware Regional Scatterplots | Slate Star Codex
If you make a scatterplot of geographic regions it can show a strong correlation even between things that aren't really related. Why? This article explains.
statistics  SlateStarCodex  via:SlateStarCodex  ScottAlexander 
april 2016 by mcherm
My Favorite Paradox — Dev Curious — Medium
Great write-up of why results may change when you break your data into sub-groups. I want to teach this to Ben.
math  statistics  via:reddit  todo 
april 2016 by mcherm
Crowdsourced research: Many hands make tight work : Nature News & Comment
They gave the same data to 30 different reasearh teams to analyze. The techniques used varied a lot, and the conclusions differed.
science  ScientificMethod  via:HackerNews  statistics 
march 2016 by mcherm
1/3 of all people in the US killed by a stranger are killed by police
I don't know that I trust the analysis or the conclusion, but at least they're trying and they're justifying their approach.
statistics  police  policeabuse  policeimmunity  via:reddit 
march 2016 by mcherm
Why Percentiles Don’t Work the Way you Think
Some analysis of performance stats for monitoring app performance.
via:reddit  statistics  programming 
december 2015 by mcherm
Tails of Great Soccer Players | Put A Number On It!
This explains why a tiny country can field a better soccer team than China. Because the national team is made up of only the tail of the distribution, and small differences in the average matter more for the tail of the distribution than large differences in the number of people. Surprisingly interesting.
statistics  via:SlateStarCodex  math 
december 2015 by mcherm
Why do so many prisoners end up back in prison? A new study says maybe they don’t.
If you count among people released during a certain date, most get re-arested. But that over-represents repeat offenders. If you count among people imprisoned, most never return.
prison  statistics  via:HackerNews  slate 
november 2015 by mcherm
Study on helmet laws and use and bicyclist injury rates
Women were safer, bike lanes were safer. But, while helmet laws made people wear helmets, wearing them did not correlate with lower injury rates.
statistics  todo  bicycle  science  via:HackerNews 
november 2015 by mcherm
The Mathematics of Paul Graham's Bias Test - Chris Stucchio
A mathematically sound test to see if there was bias in selecting a group even if you don't know the pool of applicants by looking at how well the selected group members did.
statistics  math  discrimination  via:HackerNews 
november 2015 by mcherm
A Way to Detect Bias
Here's a way to measure bias in selecting people (or whatever) without knowing who applied. Measure success. If one group (eg. women) consistently do better then they had to reach a higher bar to get selected.
statistics  PaulGraham  math  discrimination  via:HackerNews 
november 2015 by mcherm
Perceptions Of Required Ability Act As A Proxy For Actual Required Ability In Explaining The Gender Gap | Slate Star Codex
Scott Alexander looks at actual data on female to male ratios in various professions (STEM and non-STEM) and finds extremely strong evidence for the claim that differences in math scores on the GRE account for nearly 100% of the differences.
iq  statistics  brain  sociology  via:SlateStarCodex  ScottAlexander 
october 2015 by mcherm
Prescriptions, Paradoxes, and Perversities | Slate Star Codex
Really interesting analysis of medicine for depression. Patients rating drugs well is negatively correlated with doctors rating them well. The strongest predictor of how well patients will rate the is how OLD the drugs are. Are patent law policies leading to bad medicine?
medicine  depression  science  statistics  via:SlateStarCodex  patent 
july 2015 by mcherm
Beware Summary Statistics | Slate Star Codex
Summary statistics can mislead... ask to see the scatter plot.
statistics  via:SlateStarCodex  datavisualization 
july 2015 by mcherm
What crimes are people incarcerated for? - Everything you need to know about prisons - Vox
Actual statistics on what kinds of crimes people are imprisoned for in the US.
vox  via:vox  prison  statistics 
april 2015 by mcherm
Dunning-Kruger and other memes
Apparently, several pieces of good research are often mentioned and cited as if they said something different than the research actually said. Incompetents don't think they're better than the competents, money increases happiness at all levels, not just below a cutoff, and type systems don't clearly improve programming languages.
brain  psychology  science  statistics  ScientificMethod  via:HackerNews 
march 2015 by mcherm
College rape: Campus sexual assault is a serious problem. But the efforts to protect women are infringing on the civil rights of men.
Contains actual numbers on campus rapes. The "1 in 4 women" figure is completely bogus; reasonable numbers seem to be between 0.5% and 5%.
slate  gender  statistics 
february 2015 by mcherm
Should Teens Not Drive Older Cars? : Free Range Kids
Misuse of statistics: they found that lots of kids who died in car crashes were driving old cars. But they don't know how many kids drive old cars, so you can't tell if it helps or hurts. The news media ran the story "don't let kids drive old cars".
statistics  via:FreeRangeKids  journalism 
december 2014 by mcherm
Can a classifier that never says “yes” be useful?
If you are asked to predict who will do X, it may be the wrong question. The real answer may be a way to identify a subpopulation with an increased chance (still < 50%) of X.
math  statistics  via:HackerNews 
march 2014 by mcherm
Scientific method: Statistical errors : Nature News & Comment
About how "P fishing" can distort results in science, and how it could be avoided.
ScientificMethod  via:HackerNews  statistics 
february 2014 by mcherm
Time to retire standard deviation?
Standard deviation overemphasizes outliers. Averaging the deviations is closer to what people THINK standard deviation means. [This omits all mention of the natural advantages of standard deviation when your variables have a normal distribution.]
math  via:reddit  statistics 
january 2014 by mcherm
Same data, different Histograms
Displaying the same data set with VERY different-looking histograms based just on where the bucket boundaries are set. An artificial example but a very real issue.
statistics  via:HackerNews  math  datavisualization 
april 2013 by mcherm
Whom the Gods Would Destroy, They First Give Real-time Analytics
"It is BETTER to have slow analytics rather than real-time analytics because people will misuse statistics if they are too available."
via:HackerNews  statistics  monitoring 
january 2013 by mcherm
Who wrote the 'Death Note' script?
a very detailed analysis to decide if a document is fake or real
math  statistics  logic  via:HackerNews 
january 2013 by mcherm
Effective Web Experimentation as a Homo Narrans
To avoid accepting random correlations as meaningful conclusions, specify BEFORE STARTING your A-B test what it is that you are looking for.
statistics  testing  via:boingboing 
january 2013 by mcherm
Why People in Cities Walk Fast - Jobs & Economy - The Atlantic Cities
On a Log-Log graph (where ANYTHING is linear, so take this with a grain of salt) town population is linearly correlated with walking speed. The bigger the city, the faster people walk.
statistics  via:reddit  personal_net 
march 2012 by mcherm
Being descended from Confucius » Language Log
A look at how far back one needs to go before people are almost certainly the ancestor of ALL living people (or none).
via:MichaelLugo  biology  statistics  math 
february 2012 by mcherm
Is There Anything Good About Men
Women and men may have "equal" abilities but they have different motivation. Men vary more than women. There may be evolutionary reasons for this.
via:HackerNews  gender  psychology  sex  philosophy  feminism  statistics 
july 2011 by mcherm
Cracking the Scratch Lottery Code | Magazine
State lottery tickets can be "cracked", giving the ability to tell winners from losers.
wired  via:MichaelLugo  probability  statistics 
february 2011 by mcherm
Gay Sex vs. Straight Sex « OkTrends
OKCupid uses their huge online dating database to answer questions about gays.
OkCupid  via:OKCupid  gayrights  sex  statistics 
october 2010 by mcherm
Statistical significance & other A/B test pitfalls : Cennydd Bowles on user experience
Better pay attention to statistical significance before you rely too heavily on A-B testing to do your design.
design  statistics 
july 2010 by mcherm
Murky Research - Fabulous Adventures In Coding - Site Home - MSDN Blogs
A CarTalk puzzle about filling cars, and analysis of false negatives.
math  statistics  via:EricLippert 
july 2010 by mcherm
Lucia de Berk – a martyr to stupidity – Bad Science
About a woman convicted of murder when there was no evidence except some mis-applied statistics.
law  math  statistics  via:HackerNews  personal_net 
april 2010 by mcherm
The Democrats Are Doomed, or How A ‘Big Tent’ Can Be Too Big « OkTrends
Very interesting article explaining why the Republicans are much more cohesive than the Democrats.
via:OKCupid  politics  OkCupid  datamining  statistics 
april 2010 by mcherm
Anscombe's quartet - Wikipedia, the free encyclopedia
A set of 4 datasets with identical mean, variance and correlation, but DRASTICALLY different meaning. Proves you have to graph your data before you begin analyzing.
math  statistics  via:reddit  Wikipedia 
january 2010 by mcherm
Rubrics And The Bimodality Of 1-5 Ratings
When you give people a chance to enter a ranking from 1..5, it isn't very well distributed. Here is a possible solution: using other questions as leading hints to gently nudge people into using a wider variety of scores, and track both average and variance.
socialcomputing  voting  statistics 
december 2009 by mcherm
Data Coverage and Zestimate Accuracy - Zillow
How accurate is Zillow at predicting home prices?
zillow  mortgage  statistics 
november 2009 by mcherm
The Myth, the Math, the Sex - New York Times
Surveys say that men have on average more sex partners than women (about 7 to about 4). But it is mathematically impossible. The statisticians reply by saying the effect is real, but perhaps the surveys exaggerate it. I think this shows that those statisticians are incompetent.
nytimes  via:reddit  statistics  science  math 
november 2009 by mcherm
Whimsley: Netflix Prize: Was The Napoleon Dynamite Problem Solved?
Analysis says: Yes... key ideas were: (1) people's ratings vary over time... allow for this; (2) movies go up and down in popularity within the population as a whole; (3) a few movies are just harder to predict. Deal with it.
via:HackerNews  statistics  netflix 
october 2009 by mcherm
FiveThirtyEight: Politics Done Right: Seen Through Sharper Statistical Lens, Anomalies in Strategic Vision Polling Remain
Yep, this particular survey firm is definitely making up the numbers instead of doing actual polling.
statistics  politics 
october 2009 by mcherm
How Races and Religions Match in Online Dating « OkTrends
Stats from OKCupid about how people react to different races and religions. Summary: their stats are good enough to show that astrological sign makes ZERO difference; people say religion makes a big difference; people who take religion less seriously are preferred; people say race doesn't make any difference.
statistics  race  religion  via:HackerNews 
september 2009 by mcherm
FiveThirtyEight: Politics Done Right: Comparison Study: Unusual Patterns in Strategic Vision Polling Data Remain Unexplained
Someone does an analysis and points out that the data from one particular US polling company appears as if it might be made up.
math  fraud  poll  statistics  via:HackerNews 
september 2009 by mcherm
2845 ways to spin the Risk | Understanding Unce - Flash Player Installation
An example about how to lie with statistics. The exact same statistic presented thousands of different ways, with some information about how people will respond.
infographics  statistics 
august 2009 by mcherm
Public Praises Science; Scientists Fault Public, Media: Section 4: Scientists, Politics and Religion - Pew Research Center for the People & the Press
Stats show that scientists knew about the Bush administration's "War on Science", but the general population didn't. And not a lot of scientists are Republican.
science  politics  statistics  via:reddit  personal_net 
july 2009 by mcherm
What Is Ergodicity? - Individual behavior and ensembles - Softpedia
Definition of "Ergodicity". Say you take a sample from a group over a time period. If you sample many individuals over a short time it MIGHT be good, or it might be off because not all times are typical. If you sample a few individuals over a long time it might be good or it might be off because not all individuals are typical. When some aren't typical the population lacks ergodicity. Humans groups usually lack ergodicity, thus some inferences are invalid (eg: "blacks have higher crime rates thus I shouldn't trust a black" is invalid because humans don't have ergodicity.
words  math  statistics  vocabulary 
july 2009 by mcherm
Rape Fantasies and Hygiene By State « OkTrends
Extracting very interesting statistics from the data people input into an online dating site. The amount of data that they have is ENORMOUS (people are strongly motivated to answer THESE surveys).
statistics  infographics  survey  via:HackerNews  OkCupid 
june 2009 by mcherm
The Devil Is in the Digits: Evidence That Iran's Election Was Rigged - washingtonpost.com
You can analyze the low-significance digits in the vote counts in Iran's election and look for the accidents of "random" numbers that get made up by humans (who aren't very random). Results: only 1/2 of 1% chance that Iran's election could have happened this way without fraud.
politics  math  statistics  election  voting  via:reddit 
june 2009 by mcherm
Compare Languages - Ohloh
An interesting set of data on what languages are used how much.
languages  programming  statistics  via:reddit 
june 2009 by mcherm
Study: racial profiling no more effective than random screen - Ars Technica
Math says, don't screen people based on their profile... screen randomly with a bias based on the square root of their profiled probability of problems.
via:digg  statistics  math  terrorism  ArsTechnica  security 
february 2009 by mcherm
« earlier      
per page:    204080120160

Copy this bookmark:





to read